Studying Post-Editese in a Professional Context: A Pilot Study

Studying Post-Editese in a Professional Context: A Pilot Study Lise Volkart FTI/TIM, University of Geneva Switzerland lise.volkart@unige.ch Pierrette Bouillon FTI/TIM, University of Geneva Switzerland pierrette.bouillon@unige.ch Abstract The past few years have seen the multipli- cation of studies on post-editese, follow- ing the massive adoption of post-editing in professional translation workflows. These studies mainly rely on the comparison of post-edited machine translation and hu- man translation on artificial parallel cor- pora. By contrast, we investigate here post-editese on comparable corpora of au- thentic translation jobs for the language di- rection English into French. We explore commonly used scores and also proposes the use of a novel metric. Our analy- sis shows that post-edited machine trans- lation is not only lexically poorer than hu- man translation, but also less dense and less varied in terms of translation solu- tions. It also tends to be more prolific than human translation for our language direc- tion. Finally, our study highlights some of the challenges of working with comparable corpora in post-editese research. 1 Introduction Much progress has been made since the seminal paper by Baker (1993) introduced the notion of translation universals and suggesting “to capture” the differences between original and translated lan- guage using comparable electronic corpora. Cor- pora of translated texts have been widely stud- ied since then and research by Olohan and Baker (2000), Cappelle and Loock (2013) and Volan- sky et al. (2019), among others, have revealed © 2022 The authors. This article is licensed under a Creative Commons 3.0 licence, no derivative works, attribution, CC- BY-ND. the existence of translationese features. Following this, a new type of corpus-based translation stud- ies has recently emerged together with the boom of neural machine translation (NMT) systems and their large integration into professional translation workflows. Those new studies are interested in the phenomenon of machine translationese and post- editese, the latter being defined as “the expected unique characteristics of a post-edited text that set it apart from a [human] translated text” (Daems et al., 2017). Our study falls within this area of re- search and focuses of post-editese in professional context. First, we provide a short literature review of previous work on post-editese that will allow us to highlight the novel aspects of our research, as well as the common components that could consti- tute the basis for the development of a consistent methodology for the study of post-editese. Subse- quently, we present the main goals of our study, as well as our research questions. We then describe the comparable corpus used for our pilot study and discuss the main advantages and drawbacks of such a corpus for the study of post-editese. Follow- ing this, we describe the experiments conducted and results obtained. Finally, we provide a sum- mary of our findings and some perspectives for the future continuation of this work. 2 Related work This section presents some of the recent studies in- vestigating the differences between human and raw and/or post-edited machine translation output. ˇCulo and Nitzke (2016) conducted a …

Content truncated to 500 words